CAPI Data Analytics Framework

Taking a Byte Out of
Corruption
A Data Analytic Framework for Cities to Fight
Fraud, Cut Costs, and Promote Integrity
The Results of the Center for the Advancement of Public Integrity Data Analytics Working Group
February 28, 2017
TABLE OF CONTENTS
EXECUTIVE SUMMARY .................................................................................................... 2
BACKGROUND .................................................................................................................. 4
20th Century Tools for a 21st Century Change ............................................................................................. 4

Data-Driven Approaches ............................................................................................................................ 4
COMMON RISKS AND POTENTIAL APPROACHES ....................................................... 5
Fraud by Inspectors ....................................................................................................................................5

Human Resources-Related Fraud by Public Figures .................................................................................7
Benefits Fraud ............................................................................................................................................9
Campaign Finance Violations and Theft of Public Funds ....................................................................... 11
Petty Theft of Public Resources and Inventory ........................................................................................ 13
Procurement Fraud ................................................................................................................................... 15
Vendor and Contractor Fraud ................................................................................................................... 17
Fraudulent Legal Claims against the City ................................................................................................ 18
Fraud and Corruption by Elected Officials or High Level Officials ......................................................... 19
BEST PRACTICES IN DATA MANAGEMENT.................................................................20
Data Structuring ...................................................................................................................................... 20

Data Cleaning ........................................................................................................................................... 21
Creating a Data-Oriented Organizational Culture .................................................................................. 23
CAUTIONS AND CONSIDERATIONS .............................................................................25
Data Access and Privacy and Bias Issues ................................................................................................ 25

The Reliability of Results ........................................................................................................................ 26
Over-Reliance on Data ............................................................................................................................ 26
Gauging Effectiveness ............................................................................................................................. 27
POSSIBLE NEXT STEPS ....................................................................................................28
CONCLUSION ...................................................................................................................30
APPENDICES ..................................................................................................................... 31
Appendix A: Research Methodology ........................................................................................................ 31

Appendix B: Other Applications of Data Analysis .................................................................................. 33
Appendix C: Starting Points- Ten Ways to Use Existing Data to Fight Fraud ....................................... 34
1
EXECUTIVE SUMMARY
In recent years, the emerging science of data analytics has equipped law enforcement agencies and urban
policymakers with game-changing tools. Many leaders and thinkers in the public integrity community believe
such innovations could prove equally transformational for the fight against public corruption. However,
corruption control presents unique challenges that must be addressed before city watchdog agencies can harness
the power of big data. City governments need to improve data collection and management practices and
develop new models to leverage available data to better monitor corruption risks.
To bridge this gap and pave the way for a potential data breakthrough in anti-corruption oversight, the Center
for the Advancement of Public Integrity (CAPI), with the support of the Laura and John Arnold Foundation,
convened an expert working group of leading practitioners, scholars, engineers, and civil society members to
identify key issues, obstacles, and knowledge gaps, and map a path forward in this promising area. CAPI
supplemented the deliberations of this working group with further research and more than forty field interviews
in New York and Chicago. (See Appendix A for more detail on our sources.)
This yearlong study came to the following conclusions:
Data-driven investigative approaches require robust, consistent, and reliable data. City governments
should implement policies to improve data collection and management in order to undertake pilot
projects testing the usefulness of data analytics to monitor and detect potential fraud, waste, and
corruption.
Efforts to improve data collection and pioneer data-driven approaches should focus on high-priority
areas of corruption risksituations or circumstances common to city government that are particularly
prone to fraud, waste, abuse, or corruption. These areas, outlined in the report, are where cities can get
the most bang for their buck by using known fraud indicators to detect red flags among existing data
sources. While data-driven approaches will not magically identify corruption, they can help pinpoint
potential corruption risks that enable investigators to more efficiently target their efforts.
What cities can do now is to identify starting points to test data-driven efforts, using existing data
sources and known indicators. And as practitioners in various cities make headway in this fast-changing
field, CAPI can facilitate the sharing of knowledge and best practices to accelerate progress against
mutual challenges.
The following framework explores these findings in greater detail, to provide independent guidance for cities
seeking to develop data-driven investigative approaches. Neither a technical manual nor a theoretical study, the
framework aims to be practical, actionable, adaptable, and sensitive to the needs and interests of city watchdogs.
The framework has several parts:

Part I gives background information about project goals and context and delineates key concepts.
Part II details key challenges facing cities in terms of fraud, waste, abuse, and corruption for data-driven
approaches, and discusses available data sources and potential indicators that may present pathways to
address such challenges.
Part III presents best practices concerning data collection and maintenance, indispensable to any city or
agency intending to use data-driven investigative methods.
Part IV notes relevant cautions and considerations, such as foreseeable obstacles and issues of
performance measurement.
Part V concerns next steps for developmenta few illustrative examples of data-driven monitoring and
investigative methods that may be relatively easy to implement, as well as areas for future research.
Part VI concludes the framework with a final summary of key points.
2
Finally, there are three appendices:
o Appendix A summarizes the research methodology and lists the working group members who
contributed to this report and the interviewees who helped inform it.
o Appendix B discusses relevant examples of applications of data analytics in related fields.
o Appendix C is a Top Ten list of starting points for data-driven approaches to fraud and
corruption risks using readily available resources.
3
I. BACKGROUND
20th Century Tools for a 21st Century Challenge
Corruption, the abuse of public authority for private gain, poses a tremendous challenge for communities across
the country, costing U.S. taxpayers tens of billions of dollars each year and spurring a vicious cycle of citizen
distrust and disengagement.1 Recently, state and local corruption schemes have grown more sophisticated,
involving manipulation of public resources, complex bid rigging, and money laundering conspiracies. For
example, New York Citys CityTime scandal involved a massive $600 million procurement fraud related to a
city-wide payroll computerization project.2 New Jerseys Operation Bid Rig resulted in the convictions of
dozens of people for a conspiracy involving money laundering, political corruption, and even interstate organ
trafficking.3 The unprecedented scale of corruption and fraud in the Hurricane Katrina recovery efforts have
resulted in more than 1,300 indictments.4 Additional examples of data analytics are described in Appendix B.
Currently, corruption investigations predominantly rely upon random spot-checks and tips from whistleblowers.
Both strategies are time-tested but limited. Random spot-checks can be scattershot and laborious. Tip-driven
investigations are reactive and often dependent upon whistleblowers or other witness statements, which may or
may not be reliable. In theory, data-driven approaches could be more efficient, directing limited resources to
the most probable problem areas. They could be more effective, uncovering evidence of systemic corruption
risks. And they could be proactive, identifying risks before they metastasize into entrenched problems.
Unfortunately, the application of data analytics to corruption control is complex. Unlike typical uses of urban
informatics, corruption has no straightforward inputs, outputs, or metrics of success. Which data would be
relevant? What red flags should investigators look for? What benchmarks can agencies use to measure gains and
fine-tune approaches? Such thorny questions have delayed the deployment of data-driven approaches to the
fight against municipal corruption. Data-driven investigative methods used in other fields of law enforcement
have yet to be translated into effective and scalable corruption control.
Data-Driven Approaches
Information is at the heart of any investigation. Any investigation involves obtaining information and then
sorting through it to uncover facts that are relevant, significant, and credible, and to marshal those facts into an
accurate narrative. Throughout this process, leads from a variety of sources can help an investigator find,
interpret, and evaluate information.
A data-driven investigation might use data to gather evidence to supplement leads from other sources such as
complaints, witness interviews, or evidence from other investigations. Data Analytics is the process of
examining large data sets in order to generate useful leads, patterns or relationships directly from the data itself.
Types of data analytics particularly relevant to data-driven corruption investigations include:

Exploratory: used to uncover new patterns within data
Confirmatory: used to support or weaken existing hypotheses about data
Predictive: uses current and historical facts to inform hypotheses about future or unknown incidences
Quantitative: uses numerical data
Qualitative: uses non-numerical data like natural language, images, names, or addresses
Data mining: used to discover hidden patterns and relationships within large data sets
Outlier analysis: used to detect anomalies within data, such as incidences that deviate from an expected
pattern
In theory, data-driven corruption monitoring could initially involve exploratory analysis of relevant city-specific
data sets to find outliers for further evaluation. If this process were effective, the anomalies detected by the
process would provide enough useful leads to justify further investigation and evaluation.
4
II. COMMON RISKS AND POTENTIAL APPROACHES
All cities face common challenges in addressing waste, fraud, abuse, and corruption. This section highlights nine
areas of corruption risk that are recommended as starting points in the use of data-driven methods because of
their prominence in municipal corruption investigations and the availability of usable data and time-tested
indicators in these areas. For our recommendations for ten starting points from the below topics that are
relatively easy to implement with existing data sources, please see appendix C.
A. Fraud by Inspectors
What are the corruption risks?
Many regulated industries in cities rely upon periodic inspections, including daycare centers, restaurants,
healthcare providers, senior care facilities, for-hire vehicles, and waste removal services. Inspections of buildings
and construction sites represent one particular area of fraud in many cities, involving many actorsdevelopers,
contractors, expeditors, and inspectorswho interact with one another and the city over the lifespan of a
construction project, exposing multiple corruption risks. Bribery schemes involving inspectors can compromise
public safety and cost city taxpayers. For example, building developers might offer bribes to building inspectors
in exchange for helping to accelerate projects to completion notwithstanding the failure to complete critical
steps in the process. Fire inspectors might be offered bribes in exchange for overlooking expensive code
violations.
How can cities use existing data sources to better target corruption risks?
Inspection records can be a rich source of information to help anti-corruption officials identify incidences of
bribery or fraud. Records documenting inspection activities, typically maintained by municipal building
departments in the construction arena, for example, may contain common indicators of potential corruption.
We note that the ability to access the information contained in inspection records, like all of the other data
sources identified in this framework, will vary greatly depending on whether the information is available
electronically or on paper. While all cities should move towards electronic records, some will be further along
than others in that process, and the ability of a city to easily implement some of these suggestions will be based
on large part on how its information is kept.
A careful review of inspection records may reveal the following indicators of bribery or fraud:
Outliers in the number of reported code violations by inspectors, for example an inspector whose
reports of violations seems exceptionally high or low without an obvious explanation;
Missing inspection sheets and/or route sheets that may indicate skipped visits;
Inspection sheets with irregularities, for example being signed twice by the same inspector, rather than
by both an inspector and a supervisor;
A short amount of time between a failed inspection and a passing inspection of the same property,
which may indicate a suspicious acceleration of the inspection process.
In cases where inspectors work in pairs, especially when pairings are supposed to be randomly assigned,
look for recurring pairings of the same inspector teams;
Look for evidence of inspectors frequently being (allegedly) denied access to a site, which may be
serving as an excuse to close a complaint when the actual reason is a bribe;
Look for relationships between particular licensees or contractors or site owners/operators and the
individual inspector who works with themconsistent sign-offs from the same inspector across
different, ostensibly unrelated, projects could signal fraud; and
5
Examine how often a licensed private inspector is being used to determine if the frequency of jobs is
realistic, or whether it suggests exaggerations in the number of reported activities.
Hotline complaints can also point toward potential corruption hot spots. Cities that have complaint hotlines
(e.g. 311 in New York) or smartphone applications can leverage these robust data sources to identify incoming
complaints that are tied to particular inspectors.
Important strategies to consider include:
Look for evidence of complaints that were changed or downgraded, or where the language of a
complaint itself was downgraded, following its initial logging;
Identify cases in which complaints were negotiated to lower levels; and
Identify occurrences of unusually swift site visits or resolutions of complaints.
If a particular inspector, or group of inspectors, is suspected of engaging in corrupt practices, GPS data from
government issued cell phones and cars can indicate potential red flags for follow-up investigation.
During the investigation, cities should consider using the following strategies:
Use GPS data to track movement to ensure that inspectors are at reported locations at expected times
deviations from assigned routes may indicate suspicious behavior; and
Map data to find geographic trendsif one or more individuals are engaging in potentially fraudulent
behavior, track their movements to uncover other partners in crime or other projects.
Spotlight on: Fraud by Inspectors

In 2015, after a two-year investigation by the New York City Department of Investigation and the New York
County District Attorneys Office, over 50 defendants, including 11 building inspectors and five Department of
Housing Preservation and Development employees, were charged with multiple counts of bribery and fraud. The
building inspectors allegedly cleared complaints and stop work orders, and expedited inspections, allowing for
construction to continue even if the site was unsafe or in violation of building and construction codes. In
exchange, the inspectors received sports vehicles, a luxury cruise, and other payments. In one case, a
construction worker was injured on the job site, and an inspector instructed the contractors not to report the
case because it would raise questions about the sites safety. The investigation started when the Department of
Housing Preservation and Development noticed discrepancies in an internal audit.5
6
B. Human Resources-Related Fraud by Public Employees
Like any large employer, the city oversees massive systems of human resources to govern hiring, payment, and
benefits provision. Within this structure, cities run the risk of being defrauded by public employees who attempt
to exploit payment mechanisms and benefits for personal gain.
Examples of human resources-related fraud include:
Fraudulently submitted timesheets to obtain overtime payment;

Abuse or overuse of vacation time;
Falsifying mileage on cars for increased reimbursement;
Recruitment schemes that involve nepotism and conflicts of interest; and
City employees fraudulently receiving state benefits, or vice versa.
If an employee is found to be committing one of these types of fraud, they are likely to be engaging in other
forms of human resources-related fraud and should be reviewed accordingly.
Timekeeping data is the most important source of information to examine to identify human resources-
related fraud. Many cities have electronic systems for employees to log hours, or require employees to swipe in
and out of their work places. These data can be analyzed to spot unusual trends, particularly regarding the
payment of overtime.
Proven indicators of fraud to look for in timekeeping data include:
Excessive logging of overtime hours, including unusually high weekend overtime;

Overly consistent employee overtime (e.g., always the same day of the week, or the exact same number
of hours every time);
Individual employee timekeeping records that deviate from departmental normsthese can often be
identified by comparing time swipes among similarly situated employees;
Overtime taken without a formal request or supervisor approval;
Time sheets that have been revised after, or upon, supervisor approval;
Missing approval signatures from supervisors on time sheets;6
Vacation hours that do not line up with employee time swipes;
Increased labor hours with no corresponding increases in materials used or units produced; and
Relationships between specific employees and their work patterns, such as one being sick and the other
working overtime, or clusters of individuals with comparable titles across departments earning the same
overtime payments.
Cities that allow employees to submit reimbursement requests for miles driven on personal vehicles run the
risk that employees will exaggerate mileage accrued to obtain reimbursement.
Strategies for identifying fraudulent claims regarding mileage reimbursement include:
Look for individuals, agencies, or teams that are routinely taking maximum or close to maximum
mileage claims;
7
Compare mileage claims for personal cars with mileage logs of government-issued cars to identify
irregularities;
Verify claimed miles by double-checking the standard mileage calculation between given points (such as
the employees home address and work address).
GPS data on government-issued cell phones and vehicles can also indicate time-stealing and/or employee
fraud. More specifically, these datasets can help public integrity officials assess whether an employee is where he
or she should be to perform expected duties, or whether he or she might be absconding from official
responsibilities to conduct personal business on government time.
Use geofencing technology to establish a pre-set grid where employees are expected to work, in order
to receive notifications when someone has deviated from the area;
Investigate cases in which employees report a high number of broken GPS systems;
Use mobile phone metadata, including the times of calls, how long calls lasted, the numbers dialed, and
the approximate locations of calls, to confirm locations of employees; and
Identify remote logins to the departments network and the timing of such logins to determine employee
location.
Finally, onboarding paperwork can be analyzed to identify potential cases of nepotismthe preferential
treatment of relatives and friends. Using records of new hires, promotions, appointments, and contractors, anti-
corruption agents can flag instances that warrant further investigation. Strategies to consider include:
Look for an unusually high number of employees appointed by the same person;
Identify instances of newly created positions that originate from a single individual; and
Review employee names to look for commonly occurring family names, indicating that employees may
be related, and review employee addresses to look for recurring addresses, indicating that employees live
in the same household.
Spotlight on: Human Resources-Related Fraud by Public Employees

A common problem for many cities is the fraudulent use of overtime by city employees. For example, in 2014, a
former Baltimore Department of Transportation employee was indicated on theft charges for allegedly forging
signatures of his supervisor on overtime authorization forms for up to 2,227 hours, for which he received nearly
$73,000 in overtime pay, nearly double his salary.7
In another example, a 2015 audit found that employees in Los Angeles Department of Transportation collected
overtime pay totaling $3.3 million in a year. It amounted to a 263% increase in overtime pay in that division
between 2000 and 2015. The audit found that employees in the traffic paint and sign division overbilled an
average of $48,100 between 2013 and 2014. Several supervisors received as much as $70,000 in overtime pay
while one superintendents salary tripled after he claimed $155,319 in overtime.8
8
C. Benefits Fraud
Many cities manage the distribution of federal, state, and local benefits to eligible residents, including those
receiving Medicaid, cash assistance, food stamps, or other types of public assistance. In many cases, residents
receive payment cards to access their food and cash benefits. This card may be used to purchase authorized
food items at participating retailers or to make cash withdrawals from participating banks. Unfortunately,
common risks are that city employees illegally create false benefits for personal gain, or conspire with
beneficiaries committing fraud, often in collaboration with certain retailers. (Note that while fraud by
beneficiaries alone is another huge concern for cities that will benefit from data analytics tools, fraud against the
city that does not involve public officialsand therefore does not implicate corruptionis beyond the scope of
this framework.)
Information about how electronic benefit transfer (EBT) cards are used and how employees are creating
benefits cases may lead cities to identify fraudulent employee behavior. Although the program is managed by
the United States Department of Agriculture, city and state social services offices typically maintain benefits case
records that may provide insights into this type of fraudulent behavior.
To identify benefits fraud, cities can helpfully examine when and how benefits are redeemed:
Check to see if transactions from EBT cards are occurring outside the zip code, or proximate zip codes,
of the beneficiarys reported residence;
Examine EBT card transactionscard information that is manually inputted may indicate that the
individual does not physically have the EBT card and that a fraudulent purchase has occurred;
If a natural disaster code was used to justify the benefit, ensure that the disbursal of benefits related to
an actual natural disaster, particularly in timing and location;
Review the timing within which benefits are redeemedfraudulent benefits may be immediately
redeemed (e.g., at midnight on the first day benefits are active);
Search death records to identify potential usage of beneficiary activities on behalf of deceased clients;
Ensure individuals receiving disability benefits are documented to have disabilities by checking employee
rolls, business licenses with the Secretary of State, status with the Department of Health and the Social
Security Administration, and the State Department of Professional Regulation;
Map business processes to track the data flow of benefits claims and distribution, which can often
involve multiple levels of government that may each introduce potential risk; and
Monitor inventory and distribution to look for outliers such as single retailers or processing centers that
record disproportionate usage. Relatedly, check for higher than normal incidences of certain usages at
single retailers/processing centers, such as manual entry or keying-in of codes as opposed to card
swipes.
Cities can also adopt the following strategies to identify suspect behavior of city employees:
Look for employees who issue emergency benefits at an unusually high rate;
Review timestamps when a new benefits case is opened and closedif new cases are quickly closed, a
case worker may be disbursing benefits and closing the case before it is properly reviewed;
Observe whether certain benefits disbursal codes that require less documentation are used more
frequently than others;
Look at computer login data to ensure case workers are taking action on assigned cases;
9
Ensure the case worker and person auditing the case are not same person by cross-checking both log-in
and IP address information; and
Search for the use of expired benefit reason codes.
Spotlight on: Benefits Fraud

In 2015, multiple city employees in the New York City Human Resources Administration were charged with
exploiting benefits, such as rental assistance and nutrition programs typically reserved for the citys neediest
residents. One such scheme involved using fraudulent EBT cards to buy food and beverage items for resale.
Another scheme involved falsely registering individuals as landlords and giving them fraudulent rental subsidy
payments. To avoid detection, the employees used paper records to bypass computerized systems. In the end,
the investigation by local, state, and federal authorities found that these employees generated over $2.1 million
through their fraudulent schemes.9
10
D. Campaign Finance Violations and Theft of Public Funds
Campaign donations are subject to strict regulation, and must be monitored by city agencies to ensure
compliance with local donation limits. For example, in New York City, the contribution limit to a mayoral
campaign for the entire election cycle is $4,950. If there is a business relationship between a contributor and the
City, the limit is lowered to $400.10
Some cities also have public matching programs, in which public funds are given to campaigns to match
private contributions. As of 2015, at least seventeen local jurisdictions and thirteen states in the United States
made some form of public funding available for political campaigns.11 For example, the Campaign Finance
Board in New York City matches small campaign contributions at a six-to-one ratio. Recently, the city of Seattle
implemented a new program in which, instead of providing matching campaign contributions, four $25
vouchers are mailed to each voter. The voter is then free to donate the vouchers to his or her preferred
candidates.12 Regardless of the local system, public campaign funds must be spent appropriately, and unused
funds are required to be returned to the city at the conclusion of a campaign. Cities that choose to operate
public matching programs must be supported by a robust enforcement mechanism that is both tough and fair.
Actually enforcing the law in this area is crucial to incentivizing compliance and giving the public confidence
that their funds are not being wasted.
Campaign finance fraud can be committed by campaigns, donors, or both. Common types of campaign finance
fraud include:
Straw donorsthe making of surreptitious donations by illegally using another persons (the straw
donors name;
Donations made by non-residents, in jurisdictions where there are residency requirements for donors;
Reporting of fictitious contributions to defraud public matching programs; and
Improper splitting of large contributions from a single source into small donations to obtain increased
public matching funds.
For cities that have donation limits, publicly available information about campaign donations can be analyzed
to detect fraudulent activity. Primary data sources include lists of individual, corporate, and intermediary
campaign donors, which can be cross-referenced with detailed administrative datasets that contain information
about:
The addresses of city residents;

Registered lobbyists;
Vendors that do business with the city;
Tax documents showing income or financial losses; and
Discretionary spending by City Council members.
Approaches for reviewing campaign donations and associated datasets for corruption risks include:
Review donations from repeating out-of-district donor zip codes;

Determine whether there are multiple donations from the same address;
Determine if a campaign has received a high percentage of cash donations;
Flag and investigate same-day donations and contributions from first-time donors;
11
Review donor history to see if the amount given is in line with prior donations;
Check previous donations by the same donor to compare handwriting against the current donation;
Look for sequential money orders, which may suggest straw donors;
Verify that donations from corporations do not exceed the legal limit; and
Use a manual inspection to identify a high number of donor cards where the handwriting looks
similar or the cards look altered.
In cities that have public matching programs, or that otherwise provide public funding, it is also important to
examine public fund reporting. A citys campaign finance board will require supporting materials to
accompany reports regarding the use of public funds, following the issuance of matching dollars. Review of the
supporting materialsincluding leases, contracts, consulting agreements, and receiptscan reveal irregularities
indicative of corruption or fraud.
When considering documents submitted in conjunction with the use of public or matching funds, it is
important to:
Determine if the documents are complete and contain sufficient information;
Assess the timing of expenses, given knowledge about the sequence of campaign events;
Ensure each expense has a corresponding receipt and is related to an eligible campaign purpose; and
Confirm employment payments match up with the timing of the campaign.
Spotlight on: Campaign Finance Violations and Theft of Public Funds

The New York City Campaign Finance Board manages a public program that matches small campaign
donations at a six-to-one rate. In the 2013 New York City mayoral race, two members of candidate John Lius
campaign were convicted for violating campaign finance laws. To take advantage of the public matching
program, they created straw donors whom they reimbursed in order to enable the campaign to access
matching funds. Following the investigation, Liu was denied up to $3.5 million in public matching funds.13
12
E. Petty Theft of Public Resources and Inventory
On a day-to-day basis, city agencies are responsible for managing thousands of assets and transactions, all of
which are overseen by city employees with direct access to items themselves, the financial statements underlying
transactions, and other records documenting inventory. Fraud may occur when city employees steal inventory
for their personal use or resale, or collude with outside contractors to help them obtain supplies or materials at a
discount.
Inventory recordsincluding information about the acquisition, storage, and distribution of goodscan be
used to identify fraudulent activity. Per the Government Accountability Office (GAO), cities should ensure they
have detailed, accurate electronic inventory systems in place, along with proper accountability mechanisms for
recording and updating inventory data on a routine basis.14 When maintained properly, these automated
inventory management systems position city officials to leverage data analytics to identify fraud. Cities will
greatly benefit in trying to identify and prevent inventory fraud by moving to an electronic system; the need to
wade through paper documentation of inventory will make the suggestions below significantly more
challenging.
When examining inventory records, the following strategies may help public integrity officials identify
incidences of corruption:
Look for the continued purchase of items that have been declared in excess and/or unnecessary;
Look for a significant uptick in ordering, potentially indicating that employees are selling excess
inventory;
Look for abnormalities in inventory requirements, including high stock levels, and incomplete,
inaccurate, or outdated inventory records;
Examine cases in which inventory is declared lost, damaged, or broken, to verify the accuracy of such
reports;
Compare freight expensesthe cost of delivering goodsto inventory purchased to identify suspicious
patterns; excessive freight expenses relative to inventory purchased may be a sign that assets are being
diverted for personal use;
Look for suspicious patterns in goods being sold on secondary markets, such as eBayan unusually
high number of sales of a particular item the city uses may suggest fraud;
Examine sales documents to ensure that all funds are transferred appropriately and match the inventory
that is sold; and
Compare requirements for the production of a certain material to the actual number of materials
orderedfor example, assume one item X is purchased for each item Y to make product Z. If
records reflect the purchase of l000X and 500Y to produce 500Z, examine the excess purchases
of item X.
In addition to reviewing data, there are certain management techniques that are in wide use in the private sector
that cities might employ to limit the possibility of theft:
Implement an automated tracking system, such as an RFID (radio frequency identification) system, to
track inventory;
Track inventory at multiple points in the supply chaineach time that an item is moved, ensure that its
location is known; and
13
Limit the number of employees who have direct access to the inventory, including the number of
employees who have direct access to editing inventory data.
Spotlight on: Petty Theft of Public Resources and Inventory

In 2015, the Atlanta Police Department arrested a number of Watershed Management employees for stealing
city inventory. The Department of Watershed Management is a city agency responsible for providing drinking
water and wastewater services to city residents and businesses. The employees were accused of selling
Watershed equipment, such as copper, brass, and water meters, to recycling centers over a period of eight years.
The investigation into the thefts began when the City Auditors Office failed to find receipts for the missing
equipment. In addition to tightening management and security measures, including limiting the number of
individuals who can order, receive, and distribute equipment and requiring manager sign-off, Watershed
Management planned to implement a $20 million barcode system to track when equipment is moved.15
14
F. Procurement Fraud
Cities procure myriad goods and services from private contractors and nonprofit providers, ranging from
construction, to health and human services, to waste removal, to the purchasing of items like textbooks. The
vendor selection process is a particularly high-risk area for fraud. Cities should pay careful attention to how
contracts are awarded and to whom. Examples of procurement and contractor fraud include:
Collusive bidding for city contracts, which involves agreements between multiple prospective
contractors to undermine the competitive bidding process and inflate prices;16
Overcharging the city for projects via payroll manipulation and/or change orders;
Fraud related to the status and use of Disadvantaged Business Enterprises; and
Nepotism or conflicts of interest in the selection of vendors or contractorswhere decision-makers
favor certain vendors or contractors because of family connections, friendships, or, for elected officials,
because of campaign contributions previously made by the businessperson or his company.
Information about the vendors that compete for city contracts is a rich source of data for combatting
procurement fraud. Many cities have centralized databases with information about the vendors that have held
city contracts, which can be cross-referenced with other publicly available datasuch as the federal no
contract list, city and state records of sanctions against corporations and nonprofits, records of political
donors, and federal tax forms (e.g., Form 990 for nonprofit organizations)to identify corruption red flags.
To minimize the risk of fraud at the time of vendor selection, consider the following strategies:
When conducting background checks, search databases or lists of debarred vendors, such as the federal
no contract list or city or state records, to identify vendors with a suspect history;
Flag vendors that received poor performance ratings by city agencies for previous services provided;
Look for instances of fraudulent use of Disadvantaged Business Enterprises (DBE)businesses
that claim to be women, minority, disabled, or veteran-ownedby the DBEs themselves or by prime
contractors by comparing the date the company was DBE certified to the start date of the contract, and
reviewing prior contracts to verify their DBE status;
Look at the number of active contracts that a DBE hasa high number of contracts may mean they are
not completing the work themselves and passing contracts through to other organizations or
contractors;
Compare stated capacity to contract utilization rates in contract performance;
Review federal tax documents to identify organizations that are financially insolventthis may indicate
a higher risk of fraud; and
For nonprofits, review Form 990s for further information about:
o Prior government grants received, key officers, board members, and stated purpose;
o Unexplained financial losses;
o Problems with third-party audit reports;
o Reported instances of fraud; and
o Any lawsuits involving the organization (e.g., employees bringing wage and hours lawsuits).
Bids and proposals that corporations and nonprofit organizations submit to acquire contracts with city
agencies can signal potential fraudparticularly collusive bidding. Cities that have electronic procurement
systems, such as Health and Human Services (HHS) Accelerator in New York City, contain detailed and
15
historical information about bidders and awards, which, when properly analyzed, can lead corruption officials to
identify suspicious patterns in procurement.
Proven strategies for leveraging proposal data to combat collusive bidding include:
Implement software programs called cartel screens17 that help analyze bids to identify patterns
indicative of collusion, such as South Koreas Bid Rigging Indicator Analysis System.18
Look for unusual bidding patterns (e.g., bids that are identical, very close to each other, or suspiciously
far apart compared to previous tenders; or several bids that are the same percentage difference apart
from each other)these tactics are known as complementary bidding,19 and are used by firms to
ensure a particular bidder wins at an inflated price;
Identify instances where multiple bidders adjust their prices at the same time and to the same extent, or
bidders that submit identical bids for specific line items;
Look for patterns in who bids for certain contracts based on location or type of workif the same
group of firms always bids, rotating who wins, this might be a sign of collusion;
Look for identical errors in calculations or typos across multiple vendors proposals;
Flag cases where one bidder withdraws and becomes a subcontractor, or where a losing bidder becomes
a subcontractorthis could mean the losing or withdrawing bidder was promised subcontractor status
in exchange for colluding with the winning bidder;
Look for frequent change orders that increase overall contract value;
Look for unbalanced bidding, especially from incumbent bidders, where items not likely be needed are
underestimated and items likely to be used are price-inflated; and
Look for patterns of lowballinge.g., one vendor is consistently the low bidder in a certain
geographical area, or in a fixed rotation with other bidders.
Strategies for identifying nepotism or conflicts of interest in the procurement process include:
Look for connections between bidders and the city employees making procurement decisions by
identifying matching addresses, phone numbers, and zip codes, or connections on social media;
Look for connections between bidders and estimators in government who decide how much a project
or an item should costif a bidder appears to be consistently close to the independent government
estimate, the bidder may be receiving inside information;
Identify cases where a vendors Employer Identification Number matches an employees Social Security
Number;
Flag cases where numerous sole source contracts have been awarded to the same contractor, or cases
where an open bid receives only one proposal; and
Look for potential conflicts of interest, such as government employees with family members that work
for bidding companies, or city employees that serve on the board of a nonprofit that does business with
the city.
Spotlight on: Procurement Fraud

In 2016, John Bills, a former assistant transportation commissioner for the City of Chicago, was sentenced to
ten years in federal prison for his role in a scheme involving the Citys red-light camera contracts. In 2003,
Redflex Traffic Systems Inc. was awarded a contract to install red-light cameras in Chicagos intersections. In
the approximately ten years following this initial award, Bills used his influence to continuously expand
Redflexs business with the City, resulting in millions of dollars for the companys installation of additional red-
light cameras. In exchange, Redflex provided Bills with cash and other personal benefits, including meals, golf
outings, rental cars, airline tickets, hotel rooms, and other entertainment. The case was investigated by the
Inspector General for the City of Chicago with municipal, state, and federal partners.20
16
G. Vendor and Contractor Fraud
After a city contract has been awarded to a vendor, corruption risks remain, and vendor performance and
compensation should be monitored throughout the duration of the agreement. Cities run the risk of being
defrauded by vendors who submit inaccurate or misleading invoices for payment, or who collude with city
employees to receive fraudulent payments.
Invoices submitted by vendors, and subsequent payments made to these vendors, provide information that can
be analyzed to detect potential fraud. Many cities keep electronic records of invoices, and have a centralized
system to track payments issued. Following the award of a city contract, pubic officials should remain on the
lookout for the falsification or manipulation of invoices submitted for payment.
Cities can effectively review invoices with the following strategies in mind:
Flag cases of unusually high hours, especially when reported to be logged over the weekend, or contracts
that are significantly over-budget;
Verify that the requested payment is consistent with how much work the contractor is doing on an
hourly basis, and the progress being made on the project, to protect against intentional over-billing;
Guard against payroll fraud by cross-referencing the names and addresses of all contracted employees
using available information from the Department of Motor Vehicles (DMV) or Social Security
Administration (SSA)21each employee should have a unique taxpayer identification and address;
Compare requests for reimbursement for goods and materials against the standard market price;
For construction projects, look for frequently submitted change orders, especially those without the
architects approval;
For recurring material contracts (e.g., tiles), check the vendors prior contracts to confirm consistent
pricing;
Flag increases in the volume of contractor claims for reimbursement, or changes in the appearance of
claims forms;
Confirm requests for reimbursement are accompanied by supporting receipts;
Ensure payments are not being made to inactive or terminated city vendors;
Confirm segregation of duties for approval, recording and custody of assets;
Identify cases where the vendors purchase order is later than the date of the corresponding invoice; and
Investigate unusual patterns in charges for overhead or for general administrative expenses.
Spotlight on: Vendor and Contractor Fraud

In 2012, the URS Corporation, an engineering, design, and construction firm, settled with the Massachusetts
Port Authority (Massport) to resolve fraudulent invoicing during the renovation of Logan International
Airport. Lawsuits filed against URS Corporation allege that a former employee submitted hundreds of false and
inflated invoices, resulting in overpayment of more than $1.3 million. The employee in question, George
Papadopoulos, the URS project manager for the Logan renovation, had submitted monthly claims to URS for
thousands of dollars for personal reimbursements. Papadopoulos would inflate labor and other charges in his
invoices, and URS would overcharge Massport and then reimburse Papadopoulos. To settle the lawsuit, URS
Corporation paid over $3.3 million, which includes repayment of overcharges. Papadopoulos was sentenced to
state prison.22
17
H. Fraudulent Legal Claims against the City
Fraudulent legal claims occur when individuals knowingly submit a false claim for payment or compensation, or
knowingly make or use a false record or statement in pursuit of government payment. Fraudulent legal claims
typically take the form of personal injury or property damage incidents,23 in which the claimant seeks liability
compensation from the city government. Individuals may submit fraudulent claims or collude with city claims
officers to receive undeserved settlement payments. Claims officers are often overly-specialized in terms of the
subject matter of the claims they handle, which may lead to collusion between the officers and the lawyers who
file claims.
Officials seeking to identify this type of fraud should begin by examining the claims submitted and approved
for payment by the city. The office of the city auditor or comptroller often oversees the review and settlement
of these claims, and holds either electronic or paper records of each claim and the result.
Strategies for identifying fraudulent legal claims include:
Look for approval of claims at amounts right below the citys capthis could mean claims officers are
knowingly approving false claims as part of a scheme with the claimants and/or attorneys;
Compare the address and any other identifying information of the claimants and attorneys to the claims
officer, to identify matches or trends that suggest a personal relationship;
Use mapping techniques to identify the geographic areas where people most frequently file
complaintsan unusually high number of complaints in a particular region or at a specific site could
indicate fraud; and
Examine suspicious trends in casework with certain attorneys or law firms.
In addition to reviewing data, there are certain management techniques that cities might employ to limit the
possibility of fraudulent legal claims:
Rotate claims officers so they have to move around to different substantive areas, forcing them to
diversify their expertise; and
Ensure random assignment of claims to claims officers.
Spotlight on: Fraudulent Legal Claims against the City

In 2014, the District Attorney of Philadelphia announced charges against 24 individuals who allegedly submitted
false insurance claims totaling over $400,000. Over the course of seven years, the defendants staged slip and fall
accidents and claimed to have suffered injuries. Claimants were encouraged to select a location where the defect
in the sidewalk was not too noticeable. They were then instructed to go to the hospital or call an ambulance
after a staged fall, make follow-up appointments with doctors, and engage in physical therapy to bolster their
claims. 24
18
I. Fraud and Corruption by Elected Officials or High-Level Officials

Public functionaries and contractors are not the only participants in public fraud and corruption. Fraud by
elected or appointed officials can be difficult to detect and prosecute due to the status and prominent positions
of these individuals. If left undetected, however, a culture that tacitly accedes to corruption squanders public
resources and erodes community trust.
Information about an elected officials donors, discretionary spending, and personal contacts may lead to
the identification of corruption. By identifying who has donated to an elected officials campaign, examining
where the official chooses to allocate discretionary funding (in the case of city councilmembers), and
determining who has frequent contact with the officialparticularly lobbyistscities can identify cases of
potentially illicit activities, such as bribery. Note, however, that data work in this area can raise significant data
privacy and even First Amendment concerns as it relates to campaign donations, so watchdogs must know and
consider the relevant laws and regulations as they proceed.
In particular, cities can use the following strategies to identify corruption:
Review an elected officials use of discretionary funds, such as city council member itemsto identify
suspicious patterns;
Match campaign contributions data with vendor and lobbyists data;
Cross-reference individual and corporate donations to elected officials with specific success later on
(e.g., a businessman who has contributed to an official has obtained a sought-after contract or license,
potentially benefiting from the relationship); and
During a follow-up investigation, look at the data contained in the envelope information (including
timestamps, length of call, number dialed, or email recipient) of an officials phone and email contacts
during the application period for a city contractlook especially for suspicious contact between
contractors and the officials overseeing selection and procurement.
Public officials must file financial disclosure forms on a regular basis to publicly disclose personal financial
holdings, assets, and transactions, as well as information on income, gifts, and reimbursements. Evaluating these
disclosure filings may reveal fraudulent activity. In particular, anti-corruption officials may look for improper
disclosure filings, including a lack of specificity on matters subject to lobbying. In addition, lifestyle checks or
unexpectedly high income reported on financial disclosure forms may indicate illegal activity. In reviewing
disclosure forms, look specifically at information about the individuals positions and organizations, outside
employment, and income.
Spotlight on: Fraud and Corruption by High-Level Officials

New York City Councilman Ruben Wills, was arrested and charged in 2014 with multiple counts of fraud and
grand larceny. Wills was accused of stealing over $30,000 in public funds from the New York City Campaign
Finance Board and the New York State Office of Children and Family Services. In both instances, Wills used a
non-profit organization that he founded, NY 4 Life, to hide and redirect funds. In one scheme, when running
for his seat on the City Council, Wills received $11,500 in matching funds. Wills used these funds to pay a shell
company that supposedly translated and distributed campaign literature, while actually redirecting the funds to
NY 4 Life. In another scheme, NY 4 Life was awarded a contract to conduct four public service projects, but
only implemented one program, while Wills pocketed the remaining $19,000 in grant funds. Through a series of
cash withdrawals, Wills used the money from NY 4 Life to pay for extravagant shopping expenses. The Wills
case was led by the Attorney Generals Office and the Comptrollers Office of New York State.25
19
III. BEST PRACTICES IN DATA MANAGEMENT
In order to capitalize on the power of data-driven approaches to revolutionize corruption investigationsor

even to test their feasibility through pilot projectscity governments need meaningful, manageable data. A
truism in computer science states: Garbage in, garbage out. Data is only useful for computer-assisted analysis
if it is reliable, consistent, and machine-readable.
An important step in this process is to clearly define the purpose of data collection. Knowing the intended use
of a data source will help shape each stage of the process of data collection and management. And it will also
help guide the decisions of data reporters, the people who input the data.
Data Structuring
For data to be digitized and made useful for data analytics, it must be consistently structured. In the case of
qualitative data, this may require conscientious tradeoffs. Data that is more rigidly categorized is easier to input
and compare, but may oversimplify situations or fail to capture significant details or distinctions.
One important trade-off is between standardization and flexibility. Counterintuitively, even simple sets of
qualitative data can contain ambiguities that are hard to reconcile. For example, many individuals may
reasonably type in the same address in varying ways:
15 Lovers Lane, Apt. 1 A

15 Lovers Ln. #1A
15 Lovers Ln, 1a
Etc.
Discrepancies can be minimized through internal transmission standards, the use of clear guidelines and
standardized templates. Collection specifications should include data nomenclature, definitions, acceptable
enumerations, and rules on referential integrity to ensure that changes to one data set do not unintentionally
affect related sets of data. The specifications should also include rules for formatting common values such as
dates, names, addresses, and organizational names. It is crucial that reporters, custodians, and managers of data
are trained to follow these rules consistently and are given reference materials and channels to seek help or
further guidance.
End-user software design may help reduce discrepancies further. A drop-down menu of options or series of
binary yes/no questions may force users to choose from an array of acceptable inputs. An auto-complete
feature or auto-check feature may flag improperly entered data. Users may be prompted to confirm entered data
after it is translated into a standardized format, like the way that many e-commerce retailers prompt shoppers to
confirm their address. The Python programming language (discussed below) converts addresses into machine-
readable quantitative information as geocoordinates.
Data is useless without context. Thus, it is imperative to develop a data dictionary, a comprehensive directory
of information about the structure and format of data in a data management system, using industry standards
whenever possible. A data dictionary may include a catalog of codes that represent relevant information values.
For example, the medical field has long used volumes of diagnostic codes, billing codes, and other classification
systems to ensure that all medical professionals can communicate potentially critical information with the same
vocabulary. If all participants input data using a shared vocabulary and format, the data is more consistent and
machine-readable.
20
However, in some situations involving qualitative data, open fields are preferable for data entry. Forcing users to
pigeonhole data into preconceived typologies may introduce distortions or biases by shoehorning complex
information into predetermined value sets. Its often useful to include at least one open field for notes or
comments so that uncategorizable information is not lost entirely.
For some purposes, it might be useful to develop a set of standardized tags to index data. Tags are non-
hierarchical keywords or terms that serve as metadata, or information about data, making related data points
easier to index for browsing or searching. Data sets like case records, incident reports, or networks of related
individuals have trends that are easier to analyze if occurrences of select attributes or circumstances can be easily
tracked. In a large set of tagged data, machine learning may be applied predictively to find latent relationships or
trends. Quantitative tags, such as geotags (which convey geographical information) are designed for machine-
readability.
When data collection is outside of an agencys control, in the case of data supplied by outside departments, it is
especially important to track the source and provenance (or chain of custody) of each data set, to follow up on
any inconsistencies or issues that emerge in later analysis. The data management team should keep a record of
names of individuals from outside departments who supply information, as well as a master list of all computer
programs or other data sources used in cleaning or otherwise manipulating the data. This list should be regularly
updated, as data protocols may change rapidly as government operations are reorganized. Some questions to ask
in assessing data received from an outside source include:
Where did this data come from?

How was is collected?
Why was it originally collected?
What kinds of biases might affect this data?
Who is the Point of Contact for this data?
Perhaps the surest and most cost-effective way to ensure data accuracy is to make data sets public whenever
possible. Watchdogs in the media, civil society, and general public can then do their own checks and data
analysis to add extra sets of eyes to existing oversight.
Spotlight on: Open Data Standards

New York Citys Department of Information Technology and Telecommunications maintains the Open Data
Policy and Technical Standards Manual, an updated manual of technical standards and best practices to
harmonize data collection and dissemination efforts citywide. The manual is a great resource for other cities
seeking advice on issues raised above, such as reconciling addresses and building data dictionaries.
Data Cleaning
For government entities working to combat corruption and fraud through data analytics, it is crucial to ensure
the data being analyzed is both accurate and usable. Data cleaning is a process used to correct and remove
irrelevant and inaccurate records from a database. Common errors in a database include spelling errors, missing
information, invalid entries, redundant entries, and inconsistent formatting.26 Most data sets require some form
of cleaning, from reordering columns or references, to removing duplicates, to combining related sets.
The process of data cleaning can be carried out manually or automatically. Manual data cleaning is generally used
in simple cases and involves individual, line-by-line review of data entry to check for irregularities, duplications,
and omissions. In more complicated, larger data sets, automatic data cleaning can be carried out with the help of
computer software. Data cleaning software performs a variety of functions, including:
21
Standardize information within a data set;
Remove inconsistencies in data;
Identify duplicate entries;
Identify mismatched data entries;
Transform data formatting;
Identify and exclude missing fields;
Maintain archives of raw data; and
Provide a collaborative platform for users to interact with data.
Three commonly used software languages for data cleaning are Python, R, and SQL. Python and R are open-
source, which makes them especially appealing to cost-sensitive users, due to the widespread availability of free
resources such as online tutorials and documentation, adaptable modules, and developer communities. SQL,
which can be integrated with Python and R, is optimal for relational databases (such as an organizational chart
or a social network or any data stored in XML format), since it stores data relationally using metadata rather
than in fixed columns. Those languages all allow for the application of interactive commands to process data,
generating automated reports and graphics.
Below are some examples of more commonly used and well-reviewed data cleaning software programs. Before
introducing particular software options, officials should understand the specific needs of their agency and define
a clear set of goals for using the software:
Ab Initio is used for both data integration and data management. A unique feature of this software is
its Graphical Development Environment (GDE). The GDE components can be used to access data
from a variety of sources, for example, web services. This extracted data can then be transformed
and loaded to any given destination.
Datapreparator provides a variety of tools to clean and transform data sets using a graphical user
interface. The software does not store data sets on the computer hard disk and, therefore, can handle
very large data sets.
Datamartist is a tool for data examination. It can be used to combine and clean multiple large data
sources into one data set and provide instant analysis of key indicators to assess data quality.
OpenRefine is useful for transforming data formats. A significant feature of this program is that it
allows users to link and extend data sets with other web services.
Syncsort provides data cleaners with fast, high volume sorting. It can collect, integrate, and distribute
big data efficiently.
Pentaho Data Integration uses a consistent data cleansing approach that is repeatable and automated.
Regardless of the selected software, it is important to protect against the unnecessary loss of information. Prior
to deletion of duplicate or invalid data, as a practical matter, the data cleaning process should include an
opportunity to review and correct flagged entries. In addition, once a dataset has been cleaned, it is important to
preserve its integrity by preventing the introduction of new contaminated data. Original data may need to be
carefully preserved as is for legal purposes.
Finally, it should be noted that technological gains are helping to reduce the burden of data cleaning. For
example, probabilistic data matching techniques can help to link similar entities in data sets that contain errors
and inconsistencies, without undertaking more comprehensive data cleaning efforts.
22
Creating a Data-Oriented Organizational Culture
A government agency must integrate data collection, management, and usage into its organizational culture,
to ensure that decision-makers have access to rich, reliable, and usable data.
Many municipal offices still use paper-based data records. Such records are easy to misplace or misread and
costly to digitize. There are three basic options for digitization of paper records. The cheapest option is to
scan them, which produces files in formats that are not machine readable, such as portable document
format (.pdf). A more costly option is to use document reading software, which may be costly and error-
prone, especially if the records are handwritten or otherwise difficult to interpret. Finally, records may be re-
entered into a digital format, which can be labor-intensive.
Thus, it can be worth the upfront cost to create a digital system for data entry. Even better is a secure web-
based system that allows data to be shared, updated in real time, stored securely (such as in a cloud-based
system), and accessed remotely.
But even if an office adopts better data management practices for the future, legacy databases may still be in
suboptimal formats. Organizations may need to carefully consider how to integrate legacy databases with
more current sources, and how to ensure all data is accessible and properly contextualized.
Data reporters need to be included as an integral part of the data management process. They are the ones
making frontline decisions about data input that may significantly impact results downstream. It is important
that they be properly trained and informed about the purpose of data collection and given a channel to ask
questions and offer feedback, since they may be best positioned to know if questions or categories dont
accurately reflect the facts on the ground. Furthermore, their morale may suffer if they feel alienated from
the process, which can lead to inconsistent or inadequate data entry. Anecdotal evidence suggests that city
workers who understand that the data they input into a system will be used to help make decisions about
targeting corruption investigations to find misuses of taxpayer funds may be more conscientious and
resourceful than those who believe the data they input will just sit in an overlooked file, gathering real or
virtual dust. Data reporters who feel alienated from the process or its outcomes may enter data with more
inconsistencies.
End-users, such as key decision-makers, must be adequately trained and committed to data usage. Veteran
investigators or policymakers who are unfamiliar with data-driven methods may be skeptical of their
reliability and reluctant to integrate new tools into their workflow. Since data analytics work best in tandem
with traditional investigative methods, these new tools wont produce results unless continuously tested and
honed. Users need to know the purpose, advantages, and limitations of these methods, and their feedback
must be incorporated into new iterations.
Because people tend to be best at interpreting visual information, data visualization is worth investment in
design, engineering, and user experience. Without a clear, intuitive visual interface, users may overlook
important results and trends.
Technological adoption often follows a steep learning curve, with corresponding upfront costs. Data
reporters and end users need retraining, so that they are both capable and willing to deploy new methods.
Engineers and designers need to field test and refine the technology and address unanticipated problems.
Data structures need to be adapted to be more useful for data analytics, which may require re-entry, re-
formatting, reinterpretation, and new checks for accuracy and consistency. All these changes may have
second-order repercussions that require time and effort to address.
Thus, it may be easier to implement data-driven approaches as part of a broader revamp of technology,
processes, and organizational culture. New agencies or agencies that are currently in transition, or
23
contemplating overhauling internal procedures, may be better positioned to adopt new methods than more
established ones.
At its best, the usage of data analytics is an iterative process, a cycle of trial and error that leads to
continuous improvement of methods and greater accuracy and reliability of results. Hypotheses should be
tested and confirmed as new data is received. Core indicators must be routinely reviewed and reassessed.
Employees and partners need to periodically query data sources and use curiosity and creativity to expose
potential risks, biases, latent assumptions, or inefficiencies in data-driven processes.
Data reporters should be included in cycles of improvement as well, since they may be best positioned to
know if the data captured accurately reflect facts on the ground. They should be given channels to provide
feedback through an online form, e-mail address, a phone number, or smartphone application.
Increasingly, government organizations are creating data officer positions or teams to coordinate and
improve data management policies and facilitate data sharing across offices, such as New York Citys
Mayors Office of Technology and Innovation and Mayors Office of Data Analytics, or Seattles Chief
Technology Officer, or Philadelphias Chief Data Officer. A citys chief data officer, chief technology
officer, or open data coordinator can play a crucial role in harmonizing data collection and implementing
best practices. Such offices can be important partners for watchdog agencies.
24
IV. CAUTIONS AND CONSIDERATIONS
Data analytics can yield unclear and unreliable results when critical variables cant be accurately or consistently
measured. Unfortunately, these are common challenges in monitoring for corruption. Agencies implementing
data-driven approaches must be careful to consider relevant benefits and risks.
Where to House a Data Analytics Program

Careful consideration should be given to what component of city government will be responsible for
conducting corruption-related data analytics and resulting investigations. If possible, cities should consider
building internal capacity for this purpose, either within an existing city watchdog agency or otherwise, as
this will be preferable in the long term to outsourcing. The least desirable option may be contracting with a
private sector company to provide data analytics services of this sort to the city as this is likely cost-
prohibitive in the long term and raises privacy concerns, discussed further below. If data capacity is built
internally, cities must consider how success should be measured and put into place mechanisms for ensuring
that goals are being met and that there is someone watching the watchdog, so to speak.
Data Access and Privacy and Bias Issues

Unlike many other areas of urban informatics, corruption investigations often involve data that is legally
restricted due to concerns of privacy and confidentiality. Such concerns can be addressed through the
conscientious stewardship of data, attention to cyber security, and careful data aggregation and
disaggregation. After all, in tracing patterns indicative of corruption risks, individual identity may be less
relevant than the various values or relationships attached to each data point. In the private sector, data
aggregation to preserve privacy is a frequent strategy employed in areas with sensitive data such as
healthcare, consumer behavior, and metering of utilities or usage of wireless networks.
Law enforcement agencies must be careful to safeguard the data they share with partners and set boundaries
on data access or usage as necessary. Relatedly, the design of a process for data-driven investigative methods
should account for differing levels of data access among users, and the extent to which certain data sets are
access restricted. Is the data publicly available? Is it available only to certain users? Does it require a specific
request, or even a court-ordered warrant? Can the data be accessed only on a case-by-case basis or is it
available in aggregate? Legal restrictions may slow down, if not necessarily bar, usage of certain data sources,
limiting their usefulness, portability, or scalability. Arguably, caution about such restrictions contributes to
information siloing in city government, as agencies are reluctant to share data that may result in legal risks
if inadvertently exposed or deliberately leaked.
Of course, law enforcement agencies tend to have measures in place to manage confidential data and the
findings of their investigations are not intended for public access. But agencies must be careful to maintain
confidentiality of data and comply with legal standards and requirements and local ethics codes. Ultimately,
an agency that seeks to use restricted data in novel ways should seek advice and approval from relevant legal
authorities, to avoid legal risks down the road. For example, the use of warrantless geolocation information
or the monitoring of public employees may raise considerations untested in court, depending on the
jurisdiction in question and specific circumstances at issue.
Agencies will also want to be sensitive to the concerns of protected classes and groups when conducting
data analysis. Whenever historical data is used to make predictions, there is the possibility that the data
encodes past biases that can be exacerbated through analysis. Even perceptions of such bias can be
problematic, so cities should be careful to consider the lingering effects of structural discrimination when
developing their programs.
25
The Reliability of Results
Outlier analysis is only effective if there is enough reliable data to establish a baseline or control of what
values or data patterns would constitute normal. In some data sources, this can be difficult as well, due to
the lack of established benchmarks or known correlations of conditions and results. When it comes to
restaurant inspections, for example, there may a wealth of historical data. But when it comes to big-ticket
procurement contracts, the data points available may be too limited and hard to compare.
Even in systems that feature large-volume, multi-variable data sets, investigators must be skeptical and
measured and wary of over-reliance on data. False positives and misleading trends can emerge from large
data sets.
One of the greatest challenges in using data analytics is avoiding confirmation bias, the selective
interpretation of data to reinforce preexisting beliefs or support preset theories. Data can be used to test
hypotheses, or explored for new hypotheses, but interpretation must be carefully controlled to allow for
alternate explanations or additional analysis. Chains of causation can be difficult to infer with confidence,
without relying on other investigative steps or methods, and misunderstood correlations are a constant risk.
Ultimately, agencies must ensure that the right expertise is deployed to evaluate the data and approaches
used, without being biased by other objectives.
One important check used in data analytics is to ensure that results are not only verifiable, but falsifiable. In
other words, elements should be included to flag results that differ from expectations. Disconfirmation can
serve as a brake on the system, to indicate to an analyst that there may be issues with data collection,
analysis, or interpretation. Results that are consistently either confirmatory or inconclusive are a signal that
analysts hypotheses are not being properly formulated or tested to allow for rival explanations.
Over-Reliance on Data
The flipside of the previous risk is that organizations may become too reliant on data. They may set their
expectations too high for data-driven efforts, may use data as a crutch, and may allow their work to be
driven by data collectionrather than the other way around.
Like the man in the joke looking for his lost car keys under a streetlight because thats where the lighting is
best, investigative processes can err by focusing exclusively on areas with strong data, rather than the areas
of most need or with the potential for greatest impact.
For example, imagine a data-driven approach that produces strong results for detecting theft of time by city
clerical workers, because of the availability of rich time data about when such workers enter and leave
restricted facilities or access online systems. This approach could be a great program to cut down on fraud
and save the city money. However, the success of the program can lull investigators into a false sense of
complacency. Are investigators ignoring similar fraud by non-clerical workers, whose time management data
is less accessible? Are investigators cognizant of how employees might have strategically responded to
crackdowns by developing crafty new workarounds that render data misleading? It is important for those
responsible for anti-corruption monitoring to constantly evaluate and re-evaluate approaches, especially in
light of information gathered through more traditional investigative approaches.
To avoid such risks, data-driven approaches are strongest when combined with human intelligence and
shoe-leather investigations. Especially in cases in which data analytic projects are used to explore or test
hypotheses, intelligence gathered through other methods are critical checks. For example, a program that
checks for favoritism in hiring solely by looking at whether a given set of workers are neighbors or family
26
members may miss workers who were otherwise closely related, merely because such data was unavailable or
veteran investigators werent consulted about real-world cases or known risks.
Gauging Effectiveness
Data-driven approaches can entail considerable upfront costs. As a new tool, the cost-effectiveness of
analytics can be difficult to measure, especially given the difficulty of measuring or assessing many indicators
of corruption. Furthermore, since data-driven methods are designed to work in tandem with more
traditional methods, their specific impact may not always be obvious or traceable. Ideally, those methods
would provide useful leads to investigators, but if those leads include false positives, then are they truly
time-savers? And if data-driven methods are meant to improve over time through a process of iteration,
how long should it take for a pilot project to achieve cost-effectiveness?
Questions of cost-effectiveness are especially important given the limited resources available to watchdog
agencies and municipal government generally, and the frequent need for such agencies to justify their
budgets in a political climate that is often wary of bloated public-sector budgets, alarmed by escalating
information technology costs, and skeptical of the value of oversight itself. Thus, performance assessment
must be a priority for data-driven pilot projects, and project leaders and technical staff should work together
to choose specific, appropriate metrics for periodic monitoring. Such metrics might include closed cases, the
duration of open investigations, assets recovered, and qualitative assessments from data reporters and end
users.
27
V. POSSIBLE NEXT STEPS
The proposals below are intended to be immediately actionable, leveraging existing knowledge and resources, to
illuminate various potential pathways for future development.
Compile a database of corruption cases

Veteran practitioners often tap their knowledge of previous cases and successful investigations to spot trends
and useful commonalities. Some even believe in a periodic cycle of corruption waves, as criminals are
emboldened, develop workarounds to existing systems, trigger crackdowns, and then lie low. However,
historical records of past cases handled by anti-corruption authorities are often ad hoc and uncatalogued.
A comprehensive database of known municipal corruption cases within one city, region, or the whole country
would provide a rich trove of useful information that could be mined to track emerging trends and guide future
efforts. Using a data dictionary, each case could be tagged with significant indicators about the circumstances of
the case, sources of evidence and investigative methods used, or other relevant factors. Although such a
database would require an upfront investment of time and resources, it could eventually provide the bedrock for
novel applications of data analytics and testing of hypotheses about corruption trends.
Revamp vendor tracking

Currently, only a few major cities maintain comprehensive databases about vendors, with information about
existing contracts, historical performance, corporate management and ownership, and other information
relevant to corruption cases. One example is New York Citys VENDEX database, a restricted-access
manual-input system that can be unwieldly to search and to cross-reference with other data sources. The
Mayors Office of Contract Services (MOCS), which maintains VENDEX, is currently implementing a top-to-
bottom overhaul of VENDEX, to make the system web-based, real-time, fully paperless, electronically-inputted,
standardized, and fully searchable.
Fraud detection is not a primary purpose of VENDEX, but the database includes cautions noted by oversight
agencies, which routinely search VENDEX entries in their investigations into waste, fraud, and corruption.
With technical assistance and input from anti-corruption practitioners, a revamped VENDEX could pioneer a
powerful engine for procurement fraud detection. The system could potentially be made more accessible, more
reliable, easier to search, and better integrated with related data sources. Other cities could consider developing
similar systems, especially those that have already begun to move procurement processes online.
Cross-reference data on nonprofit contractors

Nonprofit vendors already disclose valuable information through their tax forms, including information on
budgeting, loans, management composition, lawsuits, and instances of fraud. In New York City, information is
also collected on such vendors through VENDEX, including performance assessment and cautions logged by
law enforcement agencies. Potentially, a program able to systematically cross-reference those databases could
help identify nonprofits that are relatively risk-pronedue to red flags such as lawsuits, operational deficits,
management turnover, records of underperformance, and other issues. Data could also be cross-referenced with
databases of political donors or discretionary spending by public officials to pinpoint nonprofits to whom they
are or were formally connected. Of course, such indicators are not dispositive, and may reveal many instances of
organizational turbulence unrelated to integrity violations. Nevertheless, a data-driven approach could help
investigators focus their efforts on higher-priority investigative targets.
28
Make public disclosure statements searchable
States and cities are making progress in moving towards paperless, web-based databases of public financial
disclosure statements by political candidates, elected and non-elected public officials, and registered lobbyists.
Such disclosures present crucial information that could be cross-referenced with other data sources to find
connections between political leaders, lobbyists, and vendors. An investment of resources and technical
expertise could allow oversight offices that collect this information to work together with local watchdog
agencies to share this information more efficiently. For example, the Joint Commission on Public Ethics in New
York State is moving towards a paperless system, yet currently provides financial disclosure forms as scanned
pdf documents which are not machine-readable. A consistent online data entry system would make the data it
collects much more usable, streamlining corruption investigations and allowing watchdogs to better track entries
and relationships that merit further investigation.
Encourage data disclosure and sharing
The more data accumulated, the more opportunities there are for watchdogs to make actionable connections.
Cities should consider establishing open data policies and pressure private companies to disclose useful data
(aggregated to protect consumer privacy). The more data is open, the more that civic hackers, civil society
activists, journalists, and concerned citizens can participate in data collection and analysis. Cities might consider
supporting hackathons or collaborating with academic institutions, private start-ups, and civil society groups
to brainstorm and test new ideas to use data analytics to fight fraud and save money. Other ideas include
holding conferences to more widely brainstorm and distribute best practices in this area, and developing an
online repository of successful examples where corruption has been detected using data analytics.
29
VI. CONCLUSION
Data analytics is a promising new field with the potential to kick-start a transformation in how cities identify
waste, fraud, and corruption. Data-driven efforts to monitor and investigate abuses of public trust can help
watchdog agencies perform their duties more efficiently, more effectively, and more proactively. Particularly in
the nine areas of common municipal corruption risks cited in this framework, cities already have some available
data sources and indicators to begin to implement data-driven approaches. However, data collection and
management processes need significant improvement and harmonization before the potential for those methods
can be fully tested. Furthermore, cities seeking to make headway in this area need to consider potential pitfalls,
such as privacy concerns and the dangers of confirmation biases and an over-reliance on data.
Cities interested in pursuing the potential gains of a data revolution in anti-corruption enforcement have a great
incentive to collaborate. Despite differences in specific data sets and circumstances, problems of waste, fraud,
and corruption echo across jurisdictions. Ultimately, cities should share data with each other and work together
on honing indicators and developing stronger and more innovative data-driven methods. CAPI aims to serve as
a knowledge hub for the public integrity community, for cities to share resources and lessons learned in the fight
against corruption, including the new battleground of data analytics. To find out more about CAPI or to join
our public integrity community, please visit our website at http://www.law.columbia.edu/public-integrity or
contact us at capi@law.columbia.edu or 212-854-6186. We welcome any comments and feedback about how we
can assist practitioners, scholars, and policymakers to promote public integrity and turn the tide in the joint
struggle against corruption in communities across the country and around the globe.
30
APPENDICES
Appendix A: Research Methodology
To investigate the applicability of data analytics to the problems of municipal corruption control, CAPI
assembled a Data Analytics Working Group (the Working Group). The Working Group met four times from
March 2016 to February 2017 at Columbia Law School in New York City, to discuss questions such as:
1. What are feasible applications of data analytics to targeting corruption?

2. What data sources would be relevant, and how can they be accessed?
3. What are useful indicators of fraud or corruption among those data?
4. What tools could be developed, purchased, or adapted to glean actionable intelligence and useful
information from those data?
5. What would be the specifications for designing such tools to cost-effectively meet agency needs?
6. What technical, logistical, or political challenges must be addressed to make those tools operational?
7. How can a pilot be developed and tested within particular jurisdictions in order to test feasibility?
8. What are realistic expectations for those tools, and by which metrics should they be evaluated?
9. How can such tools be designed to comply with applicable laws and legal standards and to address
any concerns of citizens, public servants, and policymakers?
10. How can tools be shared across jurisdictions to minimize unnecessary duplication of efforts?
Working Group Members:
Jennifer Rodgers Center for the Advancement of Public Integrity (chair)

Brian Browne - Ernst & Young
Aaron Chalfin - University of Pennsylvania and University of Chicago Crime Lab in New York City
John Curran Walden Macht & Haran LLP (formerly Stroz Friedberg)
Jacqueline Eppolito - New York City Department of Investigation
Joe Ferguson Inspector General of the City of Chicago
Stephanie Glive New York City Department of Investigation
John Kaehny - Reinvent Albany
Andrew Kalloch formerly Office of the New York City Comptroller
Gabriel Kuris - Center for the Advancement of Public Integrity
Calvin Lam - New York City Department of Investigation
Charlie Linehan formerly the New York County District Attorneys Office
Elizabeth Marcello - Reinvent Albany
Jeff Merritt New York City Mayors Office of Tech and Innovation
Pasqualino Russo Windels, Marx, Lane, and Mittendorf, LLP
Ravi Shroff New York University Center for Urban Science and Progress
Marcos Soler New York City Mayors Office for Criminal Justice
Peter Weitzman - Stroz Friedberg
Milton Yu - New York City Department of Investigation
Ryan Zirngibl - New York City Mayors Office of Data Analytics
At the suggestion of the Working Group, CAPI conducted additional semi-structured research interviews with
municipal leaders, current and former law enforcement officials, data science scholars, civil society
representatives, and private sector experts, including:
Stacy Aronowitz, Office of the New York State Attorney General

Salvador Arrona, NYC Business Integrity Commission, New York City
Patrick Blanchard, Inspector General of Cook County, Illinois
Andrew Brunsden, New York City Department of Investigation
31
Thomas Caulfield, Procurement Integrity Consulting Services
Pei Pei Cheng de Castro, New York State Joint Commission on Public Ethics
Gregory Cho, New York City Department of Investigation
Lisa Flores, Office of the Comptroller, New York City
Anthony Florio, Office of the Inspector General of the City of Chicago
Michael Flowers, Enigma.io
Ronald Goldstock, Pugh, Jones, & Johnson, P.C.
Stephen Hamilton, Inspector General for the Office of the New York State Comptroller
Noel Hidalgo, BetaNYC
Ahmed Jir, Office of the Inspector General of the City of Chicago
Dmitri Jones, Office of the Inspector General of the Metropolitan Transit Authority
Jamie Kalven, Invisible Institute
Lacey Keller, Office of the New York State Attorney General
Barry Kluger, Inspector General of the Metropolitan Transit Authority
Marjorie Landa, Office of the Comptroller, New York City
William Marback, Office of the Inspector General of the City of Chicago
Michele Mark-Levine, Office of the Comptroller, New York City
James McIsaacs, Department of Procurement Services, City of Chicago
Michael Owh, Mayors Office of Contract Services, New York City
Steven Pasichow, Deputy Inspector General/Director of Investigations for the Port Authority
of New York and New Jersey
Brian Peete, Office of the Inspector General of the City of Chicago
James Perazzo, Mayors Office of Operations, New York City
Roy Pollitt, Exiger
Richard Ponce, Department of Finance, City of Chicago
Deirdre Power, Office of the Inspector General of the Metropolitan Transit Authority
Ashley Przestrzelki, Business Integrity Commission, New York City
Jesse Schaffer, Campaign Finance Board, New York City
Nicholas Schuler, Inspector General of Chicago Public Schools
Andrew Sein, New York City Department of Investigation
Matthew Serio, Office of the Inspector General of the City of Chicago
Andrew Smyth, Data Science Institute, Columbia University
Sherrill Spatz, Inspector General for New York State Courts
Sheryl Steckler, Procurement Integrity Consulting Services
Mindy Tarlow, Mayors Office of Operations, New York City
Lise Valentine, Office of the Inspector General of the City of Chicago
Melissa Villa, Office of the Inspector General of the City of Chicago
Jennifer Way, Mayors Office of Contract Services, New York City
Eugene Wu, Data Science Institute, Columbia University
In addition, CAPI extends its appreciation to Bennett Midland, LLC, and to interns and pro bono volunteers
from Columbia Law School, for assistance with research and drafting of the framework.
32
Appendix B: Relevant Applications of Data Analytics
Data-driven approaches have already led to game-changing applications in municipal government, due to the
density of information, manageable scale, concentration of resources, range of service delivery needs, and
proximity between government and citizens. Data-driven policing has contributed to the rapid decline of crime
in most major cities. For example, the New York Police Departments CompStat program to identify and
map crime-prone hot spots and emerging trends has been adopted by dozens of cities nationwide, from
Baltimore to Los Angeles. The Manhattan District Attorneys Crimes Strategies Unit has pioneered intelligence-
driven prosecution strategies, such as sophisticated mapping of gang activity and weapons usage. The University
of Chicagos Crime Lab, which recently expanded to New York City with support from the Arnold Foundation,
uses data-driven insights to determine cost-effective interventions against violent crime. Private-sector firms are
exploring the uses of data analytics to detect latent crimes such as human trafficking, insider trading, and
insurance fraud.
In the field of procurement fraud, investigators and researchers abroad and in multilateral agencies have applied
data-driven tactics to procurement contracts over the last few years:
In 2013, the European Commission, together with Transparency International, developed the
ARACHNE data analytics software to cross-check data from various public and private institutions and
identify corruption risks, conflicts of interest, and irregularities. Data is collected from publicly available
sources, including government reports and media.
Two European anti-corruption research centers, Corruption Research Center Budapest and U4, built a
python-based database of Hungarian public procurement records to study relationships between
contracts, corporate ownership and management, and political officers. The project findings included
potential indicators of corruption risk and indicators of undue political influence. The related
Government Transparency Institute conducts innovative data-driven research into procurement and
corruption risks within the European Union.
The World Bank Groups Integrity Vice Presidency (INT) has partnered with the Data Science for
Social Good (DSSG) program at the University of Chicago to preemptively avoid procurement fraud by
analyzing bids for Bank-financed projects for suspicious patterns and behavior among competing firms.
Brazils Office of the Comptroller General (CGU) recently established a permanent office, the Public
Spending Observatory (ODP), dedicated to using emerging data analytics techniques to prevent waste
and abuse in public expenditures.
Domestically, data analytics has been applied by public and private-sector organizations to detect tax fraud and
public benefits fraud. For example:
At least 20 states use, or are developing, data-driven predictive approaches to target fraud in
unemployment insurance programs. New Mexico, for example, used data analytics to cut unemployment
fraud by 60% and is now testing a program that combines data analytics with behavioral economics.27
Various states and municipalities are employing emergent data analytics techniques to catch tax cheats.28
Data analytic techniques have been applied to review large healthcare claims and billing information to
target fraud indicators and catch billing errors, up-coding, and ghost patients.29
Private companies like Paypal, FICO, and IBM have applied machine learning to fight fraud by
identifying anomalous transactions using a nonlinear network of algorithms.30
Palantir Technologies, Inc. has applied data analytics to forms of fraud including insider trading, fraud,
tax fraud, tax evasion, and synthetic identity fraud, including for clients in the law enforcement field.
Apache Hadoop produces machine learning software used to detect fraud in banking and healthcare.
33
Appendix C: Starting Points Ten Ways to Use Existing Data to Fight Fraud
Of the many indicators of fraud discussed in this report, we recommend considering the following as a starting
point for data-driven approaches. The below Top Ten approaches are relatively easy to implement and likely
to unearth corruption more readily than some of the more complicated methods described in the body of the
report.
Fraud by Inspectors
Perform an outlier analysis to identify inspectors with the highest and lowest tallies of reported code
violations.
Search for code violations that were changed or downgraded following their initial logging.
Human Resources-Related Fraud by Public Employees

Perform an outlier analysis to identify the employees who report the most overtime hours, including
unusually high weekend overtime.
Search for suspiciously consistent employee overtime claims (e.g., always the same day of the week, or
the exact same number of hours every time).
Benefits Fraud
Search for transactions from Electronic Benefit Transfer (EBT) cards outside the zip code of the
beneficiarys reported residence.
Perform an outlier analysis to identify those employees who issue emergency benefits at the highest
rates.
Campaign Finance Violations and Theft of Public Funds

Perform an outlier analysis to identify campaign donations from out-of-district donor zip codes.
Petty Theft of Public Resources and Inventory

Search for significant upticks in ordering, potentially indicating employees are selling excess inventory.
Procurement Fraud
Implement cartel screen software programs to help identify collusive bidding.
Fraudulent Legal Claims against the City

Search for approved claims at amounts right below the citys cap to identify possible knowing approval
of false claims.
34
Endnotes
1
Mikesell, John. Study: Corruption Increases and Distorts Spending by the U.S. State. Indiana University
Bloomington, 2014.
http://news.indiana.edu/releases/iu/2014/06/cost-of-corruption-paper.shtml.
2
Weiser, Benjamin. Three Contractors Sentenced to 20 Years in CityTime Corruption Case. New York Times,
2014.
http://www.nytimes.com/2014/04/29/nyregion/three-men-sentenced-to-20-years-in-citytime-scheme.html.
3
Sherman, Ted. NJ Corruption Case Finally Ends. NJ.com, 2014.
http://www.nj.com/news/index.ssf/2014/10/landmark_nj_corruption_case_ends_5_years_after_the_fbi_sting_op
eration_came_to_light.html.
4
Katrina: Four Years Later. The FBI, 2009.
https://www.fbi.gov/news/stories/2009/september/katrina_090109.
5
Manhattan DAs Office, DOI, NYPD Announce Arrests and Criminal Charges in Widespread Bribery Schemes
Involving DOB and HPD Employees. District Attorneys Office, 2015.
http://manhattanda.org/press-release/manhattan-da%E2%80%99s-office-doi-nypd-announce-arrests-and-
criminal-charges-widespread-briber.
6
Fraud Case Study- Timesheet Fraud.
http://www.dodig.mil/resources/fraud/pdfs/FraudCaseStudy_TimesheetFraud.pdf.
7
Wenger, Yvonne. Former City DOT Worker Indicted in Alleged Overtime Scheme. The Baltimore Sun, 2014.
http://www.baltimoresun.com/news/maryland/bs-md-ci-alleged-overtime-fraud-20141106-story.html; City
Employee Indicted For Fraudulent Overtime Scheme. Office of Inspector General, 2014. http://inspector-
general.baltimorecity.gov/news/inspector-general-press-releases/2014-11-16-city-employee-indicted-fraudulent-
overtime-scheme.
8
Department of Transportation Audit on Overtime. Los Angeles Times, 2015.
http://documents.latimes.com/department-transportation-audit-overtime/.
9
New York City Human Resources Administration Supervisor Pleads Guilty to Defrauding Two Public
Assistance Programs of More than $1.8 Million. The United States Department of Justice, 2016.
https://www.justice.gov/usao-sdny/pr/new-york-city-human-resources-administration-supervisor-pleads-guilty-
defrauding-two; Muller, Benjamin and Nikita Stewart. Two New York City Workers Charged With Stealing $2.1
Million from Benefits Programs. The New York Times, 2015.
https://www.nytimes.com/2015/12/02/nyregion/employees-arrested-in-scheme-to-defraud-human-resources-
administration.html.
10
Limits and Thresholds 2017 Citywide Elections. New York City Campaign Finance Board.
http://www.nyccfb.info/candidate-services/limits-thresholds/2017.
11
Malbin, Michael. Citizen Funding for Elections. The Campaign Finance Institute, 2015.
http://www.cfinst.org/pdf/books-reports/CFI_CitizenFundingforElections.pdf.
12
Berman, Russell. Seattles Experiment With Campaign Funding. The Atlantic, 2015.
https://www.theatlantic.com/politics/archive/2015/11/seattle-experiments-with-campaign-funding/415026/.
13
Former Campaign Treasurer and Fundraiser Found Guilty in Manhattan Federal Court of Campaign Finance
Fraud. The United States Department of Justice, 2013.
https://www.justice.gov/usao-sdny/pr/former-campaign-treasurer-and-fundraiser-found-guilty-manhattan-
federal-court-campaign.
14
Best Practices in Achieving Consistent, Accurate Physical Counts of Inventory and Related Property. United
States General Accounting Office, 2002
http://www.gao.gov/new.items/d02447g.pdf.
15
Leslie, Katie. Four More Arrests at Watershed. AJC.com, 2015. http://www.ajc.com/news/local-govt--
politics/four-more-arrests-watershed/3jdMEXbDiABNycKF6mfKqJ/; Performance Audit: Department of
Watershed Management Inventory Management. City of Atlanta Auditors Office, 2014.
http://www.atlaudit.org/uploads/3/9/5/8/39584481/dwm_inventory_management.pdf.
16
Guide to Combating Corruption and Fraud in Development Projects, 2017
http://guide.iacrc.org/potential-scheme-collusive-bidding/.
17
Messick, Rick. Procurement Agencies: Put in Screens!. The Global Anticorruption Blog, 2017.
https://globalanticorruptionblog.com/2017/01/11/procurement-agencies-put-in-screens/.
18
Fighting Bid Rigging in Public Procurement. OECD, 2017.
https://www.oecd.org/daf/competition/Fighting-Bid-Rigging-in-Public-Procurement-%202016-report.pdf.
19
Guide to Combating Corruption and Fraud in Development Projects, 2017.
http://guide.iacrc.org/potential-scheme-collusive-bidding/.
20
Former City of Chicago Transportation Official Sentenced to Ten Years for Corruption in Awarding of Red-
Light Camera Contracts. The United States Department of Justice, 2016. https://www.justice.gov/usao-
35
ndil/pr/former-city-chicago-transportation-official-sentenced-ten-years-corruption-awarding-red; Examples of
Public Corruption Investigations- Fiscal Year 2016. IRS, 2016. https://www.irs.gov/uac/examples-of-public-
corruption-investigations-fiscal-year-2016.
21
Marasco, Jim. Payroll Fraud: How Its Done, How to Prevent It. Stonebridge, 2007.
https://stonebridgebp.com/library/uncategorized/payroll-fraud-how-its-done-how-to-prevent-it/.
22
Construction Company Pays $2.3 Million to Resolve Fraudulent Overbilling in the Renovation of Logan
Airport. Mass.gov, 2012.
http://www.mass.gov/ago/news-and-updates/press-releases/2012/2012-01-19-urs-settlement.html.
23
http://www1.nyc.gov/nyc-resources/service/1654/file-claim-against-the-city.
24
46 People Charged in Elaborate Slip and Fall Scheme. Philadelphia Office of the District Attorney, 2014.
https://phillyda.wordpress.com/2014/04/24/46-people-charged-in-elaborate-slip-and-fall-scheme/.
25
A.G. Schneiderman and Comptroller DiNapoli Announce Indictment of NYC Councilman Ruben Wills in
Public Corruption Scheme. New York State Office of the Attorney General, 2014.
https://ag.ny.gov/press-release/ag-schneiderman-comptroller-dinapoli-announce-indictment-nyc-councilman-
ruben-wills.
26
Rahm, Erhard and Hong Hai Do. Data Cleaning: Problems and Current Approaches. University of Leipzig,
Germany.
http://betterevaluation.org/sites/default/files/data_cleaning.pdf.
27
Construction Company Pays $2.3 Million to Resolve Fraudulent Overbilling in the Renovation of Logan
Airport. Mass.gov, 2012.
http://www.mass.gov/ago/news-and-updates/press-releases/2012/2012-01-19-urs-settlement.html.
28
Newcombe, Tod. States Use Big Data to Nab Tax Fraudsters. Governing, 2015.
http://www.governing.com/columns/tech-talk/gov-states-big-data-tax-fraud.html.
29
http://www.govhealthit.com/news/part-3-9-fraud-and-abuse-areas-big-data-can-target.
30
Knorr, Eric. How Paypall beats the bad guys with machine learning. InfoWorld, 2015.
http://www.infoworld.com/article/2907877/machine-learning/how-paypal-reduces-fraud-with-machine-
learning.html; Nash, Kim. FICO Takes Fraud- Detection Techniques to Cybersecurity. The Wall Street Journal,
2016. http://blogs.wsj.com/cio/2016/02/17/fico-takes-fraud-detection-techniques-to-cybersecurity/;
Goldschmidt, Yaara. Using Machine Learning to Stream Computing to Detect Financial Fraud. IBM Research.
https://www.research.ibm.com/foiling-financial-fraud.shtml.
2017. This publication is covered by the Creative Commons Attribution-No Derivs-NonCommercial license (see
http://creativecommons.org). It may be reproduced in its entirety as long as the Center for the Advancement of Public Integrity
at Columbia Law School is credited, a link to the Centers web page is provided, and no charge is imposed. The paper may not be
reproduced in part or in altered form, or if a fee is charged, without the Centers permission. Please let the Center know if you
reprint. The views expressed here do not necessarily represent the views of the Laura and John Arnold Foundation or Columbia
Law School.
Cover Design by Freepik
36

CAPI Data Analytics Framework

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CAPI Data Analytics Framework

Uploaded by

Copyright:

Available Formats

Taking a Byte Out of

20th Century Tools for a 21st Century Change ............................................................................................. 4

COMMON RISKS AND POTENTIAL APPROACHES ....................................................... 5

Fraud by Inspectors ....................................................................................................................................5

BEST PRACTICES IN DATA MANAGEMENT.................................................................20

Data Structuring ...................................................................................................................................... 20

CAUTIONS AND CONSIDERATIONS .............................................................................25

Data Access and Privacy and Bias Issues ................................................................................................ 25

POSSIBLE NEXT STEPS ....................................................................................................28

Appendix A: Research Methodology ........................................................................................................ 31

This yearlong study came to the following conclusions:

The framework has several parts:

Types of data analytics particularly relevant to data-driven corruption investigations include:

Important strategies to consider include:

Spotlight on: Fraud by Inspectors

Examples of human resources-related fraud include:

Fraudulently submitted timesheets to obtain overtime payment;

Proven indicators of fraud to look for in timekeeping data include:

Excessive logging of overtime hours, including unusually high weekend overtime;

Strategies for identifying fraudulent claims regarding mileage reimbursement include:

Spotlight on: Human Resources-Related Fraud by Public Employees

Spotlight on: Benefits Fraud

The addresses of city residents;

Review donations from repeating out-of-district donor zip codes;

Spotlight on: Campaign Finance Violations and Theft of Public Funds

Spotlight on: Petty Theft of Public Resources and Inventory

Spotlight on: Procurement Fraud

Spotlight on: Vendor and Contractor Fraud

Strategies for identifying fraudulent legal claims include:

Spotlight on: Fraudulent Legal Claims against the City

What are the corruption risks?

In particular, cities can use the following strategies to identify corruption:

Spotlight on: Fraud and Corruption by High-Level Officials

In order to capitalize on the power of data-driven approaches to revolutionize corruption investigationsor

15 Lovers Lane, Apt. 1 A

Where did this data come from?

Spotlight on: Open Data Standards

Where to House a Data Analytics Program

Data Access and Privacy and Bias Issues

Compile a database of corruption cases

Revamp vendor tracking

Cross-reference data on nonprofit contractors

Encourage data disclosure and sharing

1. What are feasible applications of data analytics to targeting corruption?

Working Group Members:

Jennifer Rodgers Center for the Advancement of Public Integrity (chair)

Stacy Aronowitz, Office of the New York State Attorney General

Human Resources-Related Fraud by Public Employees

Campaign Finance Violations and Theft of Public Funds

Petty Theft of Public Resources and Inventory

Fraudulent Legal Claims against the City

Cover Design by Freepik

You might also like